Using coding agents with open models
summary
Claims that open models now sit close to proprietary ones on benchmarks while costing far less and that a hosted service (Token Factory) adds aggressive caching, spec decoding, and zero data retention to make inference practical. Demos show VS Code agents (Client, Cursor, Claude Code via a Token Factory proxy) calling Kimi 2.6 and others to read a repo, plan tasks, run edits, and report costs and token usage, and the setup works end to end.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:50Navis Token Factory setup
- 2:53Open models for coding agents
- 7:46Shift to coding agents
- 9:10Coding agents and models
- 10:32Misconception about latest models
- 11:33Benchmarking models and agents
- 13:43Open vs proprietary models
- 14:45Open models and capabilities
- 15:50Pairing agents with models
- 16:33Client IDE configuration
- 18:41Cursor IDE configuration
- 20:06Claude Code setup
- 23:59Mixing models for tasks
- 25:44Agent instructions best practices
- 28:08Q&A
- 34:42Closing
related talks
22mSponsored Demo: Cut the Bill, Not the Brains. AI Coding Agents That Ship More and...
25mState of Open Source AI Coding Models
14mReliable Agentic AI, the Open Way
13mEmpowering Developers through Open-Source AI
19mNo Token Budget: Agentic Software Factories on Kubernetes
37mCode is cheap. Now what?
06mYour Agent Is Wasting Tokens and You Don't Know It
31mAI Coding Agents Don't Need Bigger Models. They Need Better Tools.